Access Feature Extractor in Object Detection Model Zoo

So I try to get pretrained model
ssd = model_zoo.get_model('ssd_512_mobilenet1.0_voc, pretrained=True)`,
From the summary, SSD have:

SSD(
(features): FeatureExpander(
<Symbol group [ssd5_mobilenet0_relu22_fwd, ssd5_mobilenet0_relu26_fwd, ssd5_expand_relu0, ssd5_expand_relu1, ssd5_expand_relu2, ssd5_expand_relu3]> : 1 -> 6
)
(class_predictors): HybridSequential(
(0): ConvPredictor(
(predictor): Conv2D(512 -> 84, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(1): ConvPredictor(
(predictor): Conv2D(1024 -> 126, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(2): ConvPredictor(
(predictor): Conv2D(512 -> 126, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(3): ConvPredictor(
(predictor): Conv2D(512 -> 126, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(4): ConvPredictor(
(predictor): Conv2D(256 -> 84, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(5): ConvPredictor(
(predictor): Conv2D(256 -> 84, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
)
(box_predictors): HybridSequential(
(0): ConvPredictor(
(predictor): Conv2D(512 -> 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(1): ConvPredictor(
(predictor): Conv2D(1024 -> 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(2): ConvPredictor(
(predictor): Conv2D(512 -> 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(3): ConvPredictor(
(predictor): Conv2D(512 -> 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(4): ConvPredictor(
(predictor): Conv2D(256 -> 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
(5): ConvPredictor(
(predictor): Conv2D(256 -> 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
)
(anchor_generators): HybridSequential(
(0): SSDAnchorGenerator(
)
(1): SSDAnchorGenerator(
)
(2): SSDAnchorGenerator(
)
(3): SSDAnchorGenerator(
)
(4): SSDAnchorGenerator(
)
(5): SSDAnchorGenerator(
)
)
(bbox_decoder): NormalizedBoxCenterDecoder(
)
(cls_decoder): MultiPerClassDecoder(
)
)

I want to extract only the feature extractor of ssd (which is mobilenet).
From ssd.features, it supposed to be the mobilenet layer, but why it does not convey the mobilenet architecture? I compare from
model_zoo.get_model('mobilenet1.0', pretrained=True)
which have:

MobileNet(
(features): HybridSequential(
(0): Conv2D(3 -> 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=32)
(2): Activation(relu)
(3): Conv2D(1 -> 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)
(4): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=32)
(5): Activation(relu)
(6): Conv2D(32 -> 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(7): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=64)
(8): Activation(relu)
(9): Conv2D(1 -> 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=64, bias=False)
(10): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=64)
(11): Activation(relu)
(12): Conv2D(64 -> 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(13): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=128)
(14): Activation(relu)
(15): Conv2D(1 -> 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=128, bias=False)
(16): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=128)
(17): Activation(relu)
(18): Conv2D(128 -> 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(19): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=128)
(20): Activation(relu)
(21): Conv2D(1 -> 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=128, bias=False)
(22): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=128)
(23): Activation(relu)
(24): Conv2D(128 -> 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(25): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=256)
(26): Activation(relu)
(27): Conv2D(1 -> 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256, bias=False)
(28): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=256)
(29): Activation(relu)
(30): Conv2D(256 -> 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(31): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=256)
(32): Activation(relu)
(33): Conv2D(1 -> 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=256, bias=False)
(34): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=256)
(35): Activation(relu)
(36): Conv2D(256 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(37): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(38): Activation(relu)
(39): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)
(40): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(41): Activation(relu)
(42): Conv2D(512 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(43): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(44): Activation(relu)
(45): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)
(46): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(47): Activation(relu)
(48): Conv2D(512 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(49): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(50): Activation(relu)
(51): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)
(52): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(53): Activation(relu)
(54): Conv2D(512 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(55): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(56): Activation(relu)
(57): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)
(58): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(59): Activation(relu)
(60): Conv2D(512 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(61): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(62): Activation(relu)
(63): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512, bias=False)
(64): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(65): Activation(relu)
(66): Conv2D(512 -> 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(67): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(68): Activation(relu)
(69): Conv2D(1 -> 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=512, bias=False)
(70): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=512)
(71): Activation(relu)
(72): Conv2D(512 -> 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(73): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=1024)
(74): Activation(relu)
(75): Conv2D(1 -> 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1024, bias=False)
(76): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=1024)
(77): Activation(relu)
(78): Conv2D(1024 -> 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(79): BatchNorm(axis=1, eps=1e-05, momentum=0.9, fix_gamma=False, use_global_stats=False, in_channels=1024)
(80): Activation(relu)
(81): GlobalAvgPool2D(size=(1, 1), stride=(1, 1), padding=(0, 0), ceil_mode=True, global_pool=True, pool_type=avg, layout=NCHW)
(82): Flatten
)
(output): Dense(1024 -> 1000, linear)
)

How to get the mobilenet layer from object detection model zoo ssd? Thank you.